I built the authorization primitive AI agents are missing — here's what a week of building in public taught me

A week ago I posted a question on Reddit asking how people cryptographically prove what an AI agent was authorized to do.
I had an idea. I had no code. I wanted to see if the problem was real before I built anything.

The thread got technical fast. A senior developer called it vibecoded junk and asked five hard questions. I answered them. He asked five more. By the end he was stress testing the design instead of dismissing it. That told me the problem was real.
So I built it.